IZA DP No. 1802

Discrimination as a Competitive Device:The Case of Local Television News

Caitlin Knowles Myers

DI

SC

US

SI

ON

PA

PE

R S

ER

IE

S

Forschungsinstitutzur Zukunft der ArbeitInstitute for the Studyof Labor

October 2005

Discrimination as a Competitive Device:

The Case of Local Television News

Caitlin Knowles Myers Middlebury College

and IZA Bonn

Discussion Paper No. 1802 October 2005

IZA

P.O. Box 7240 53072 Bonn

Germany

Phone: +49-228-3894-0 Fax: +49-228-3894-180

Email: iza@iza.org

Any opinions expressed here are those of the author(s) and not those of the institute. Research disseminated by IZA may include views on policy, but the institute itself takes no institutional policy positions. The Institute for the Study of Labor (IZA) in Bonn is a local and virtual international research center and a place of communication between science, politics and business. IZA is an independent nonprofit company supported by Deutsche Post World Net. The center is associated with the University of Bonn and offers a stimulating research environment through its research networks, research support, and visitors and doctoral programs. IZA engages in (i) original and internationally competitive research in all fields of labor economics, (ii) development of policy concepts, and (iii) dissemination of research results and concepts to the interested public. IZA Discussion Papers often represent preliminary work and are circulated to encourage discussion. Citation of such a paper should account for its provisional character. A revised version may be available directly from the author.

IZA Discussion Paper No. 1802 October 2005

ABSTRACT

Discrimination as a Competitive Device: The Case of Local Television News*

Local news offers a unique look not only at customer preferences but also at the strategic response of firms to these preferences. This paper uses a combination of ratings data and newly gathered information on television stations in 25 U.S. markets to examine the decisions of competing firms and how customers respond to the journalists who appear on-air at the different stations in a market. The results indicate that there is a negative correlation between the racial, gender, and age composition of competing firms. Moreover, the ratings data suggest that the stations with relatively few blacks on-air are catering to the more discriminatory customers. While a similar result is found for age and gender, the reverse holds for other groups, suggesting possible tastes for diversity for Hispanics and Asians. Taken as a whole, the evidence supports a theoretical model in which firms differentiate via the characteristics of their employees in response to customer prejudice. JEL Classification: J71 Keywords: economics of gender and minorities, customer discrimination, product

differentiation, Nielsen ratings Corresponding author: Caitlin Knowles Myers Department of Economics Middlebury College Middlebury, VT 05753 USA Email: cmyers@middlebury.edu

*I wish to thank Daniel Hamermesh, Steve Trejo, Gerald Oettinger, Paul Wilson, Thomas Wiseman, Randall Watson, and McKinley Blackburn for many helpful suggestions. I also received numerous helpful comments from seminar participants at the University of Texas at Austin, Middlebury College, the University of South Carolina, Kenyon College, Tulane University, and Georgia State University. I am also grateful to Divya Vernenker whose help with data collection went far beyond the call of friendship.

1 Introduction

Local television news does not seem to fit the mold that we economists have

cast for customer discrimination. As Becker (1957) first demonstrated, the

racial preferences of customers can directly affect the marginal revenue product

of labor of different groups and, hence, labor market outcomes. But if, as

several recent papers have suggested (e.g., Kanazawa and Funk, 2001; Holzer and

Ihlanfeldt, 1998; Burdekin and Idson, 1991; Nardinelli and Simon, 1990; Kahn

and Sherer, 1988), consumers prefer not to interact with minority employees,

then why do we see so many blacks, Asians, and Hispanics on the local news?

In 2002 an average of 21 percent of broadcast news employees at local television

stations were minorities versus 12 percent of newspaper journalists and 8 percent

of radio broadcast employees (Papper, 2003; of Newspaper Editors , ASNE).

Given the frequent supposition of prejudice against minorities, it seems strange,

at first glance, that minorities have greater representation in the more visible

media. Might it be the case that customers actually have a preference for

diversity in some circumstances? Or are other factors at play here?

Casual observations of diversity are not the only source of interest in the

market for local television news. Identifying the presence and extent of cus-

tomer discrimination is not an easy task; it requires either directly or indirectly

finding a way to measure the preferences of different labor market agents and

how these attitudes affect labor market outcomes. As a result, most studies

of customer discrimination have focused on professional sports, where worker

output and customer demand are easily observable. The evidence from these

studies has varied considerably with the particular sport, time period, and type

of position examined. Gwartney and Haworth (1974) find that black players

increased attendance at baseball games in the 1950s; Sommers and Quinton

(1982) find that blacks had an insignificant effect on baseball team revenue in

2

the 1970s; and Nardinelli and Simon (1990) find that baseball cards picturing

minority players sell for less than those of white players. Studies of basketball

have tended to find evidence of discrimination (e.g., Kahn and Sherer, 1988;

Burdekin and Idson, 1991; Kanazawa and Funk, 2001) with the exception of

trading cards for players from the 1970s (Stone and Warren, 1999). Looking at

football quarterbacks, Arcidiacono et al. (2004) find evidence of customer tastes

for diversity. The disparities in the empirical literature could indicate that the

degree and magnitude of customer discrimination is affected by the visibility of

employees and the racial composition of customers and/or employees. Recent

studies of markets in which a large percentage of employees are black tend to find

evidence of discrimination, while studies in which blacks are not as prevalent or

not as visible are less likely to find evidence of discrimination.

While they have provided a great deal of empirical evidence, sports markets

may have more than their share of idiosyncracies. Most markets have no more

than one team in any one sport, precluding analysis of the relationship between

the characteristics of firms within a market. Moreover, the consumers of sports

are predominantly male and may not be representative of consumers in general.1

Holzer and Ihlanfeldt (1998) avoid this problem by using special survey data

for businesses in four US cities to match customer characteristics with labor

market outcomes. However, their conclusions are based on estimates made by

the owners of firms about the characteristics of their customer base rather than

on a direct indicator of customer preferences.

Television news presents another window into customer discrimination, both

because employees are visible to customers and because it offers a measure of

customer preferences through television ratings. In this paper, I use a combi-1ESPN, for instance, reports that while 45 percent of TV viewers are male, 77 percent of

its viewers are (2004). Direct comparison to local television news viewers is difficult becausethese audience profiles are not made readily available. However, a Pew Research Centersurvey found that 61 percent of female respondents and 56 percent of male respondents reportregularly watching local television news, suggesting a more balanced audience (2004) .

3

nation of ratings from Nielsen Media Research for November 2003 broadcasts

of local television news in 25 U.S. cities and data on the demographic char-

acteristics of on-air personalities. Because there is evidence of sorting among

stations within a market, with some having a much larger number of minorities

on their newscasts than others, I present a theory demonstrating that customer

discrimination can cause intra-market segregation in which firms select their

racial compositions to cater to certain groups of customers. Then, turning to

the empirical evidence, I examine how the characteristics of station employees

are related to the composition of competing stations as well as of the market

in which it is located, and whether the relationships vary with changes in laws

governing the employment of minorities in broadcast news. Also, using fixed

effects estimators, I examine how the racial make-up of a station’s on-air staff

affects ratings in order to identify and measure customer preferences for different

characteristics including not only race and ethnicity, but also sex and age.

The remainder of this chapter is organized as follows. Section 2 describes

the data. Section 3 presents a theoretical model that demonstrates how cus-

tomer preferences can lead to intra-market segregation. Section 4 presents the

econometric model and describes the results. I conclude in Section 5 with a

summary of the findings.

2 Local television news data

The data used in this study are a combination of local ratings and market

demographics furnished by Nielsen Media Research and station and broadcast

data compiled from a variety of sources. The data can be broken down into

three basic groups: ratings data, biographical data, and station data.

Ratings Data

The ratings data are collected by Nielsen Media Research using electronic meters

4

that measure whether a household is tuned in to a particular program minute-

by-minute. Nielsen Media Research aggregates the data to calculate the average

per-minute audience for a show which, in turn, is used to calculate a program’s

rating and share. The rating measures the ratio of the average number of

households viewing a program to the number of households that have television

sets (the potential audience). The share measures the ratio of the number of

households viewing a program to the number of households viewing television at

that time. The ratings data used in this paper were compiled by Nielsen Media

Research for the 25 largest “Designated Market Areas” (DMAs) in the United

States.2 They measure the average audience over the month of November 2003

for the different daily local television news broadcasts on FOX, CBS, ABC,

and NBC affiliates in each of the 25 DMAs. Ratings data are used for original

newscasts between 5 a.m. and 12 a.m. local time. Table 1 reports average

ratings and shares for local newscasts in the sample by time of day and part of

week. News broadcasts that begin in the late evening are the most popular, with

an average of 8 percent of T.V. households and 14 percent of households that

are watching television watching a particular news program at this time. These

averages are for one time of day and one station alone, and most cities have at

least 4 stations broadcasting local news in English. Taken together, the ratings

suggest that a large portion of the population watches local television news

broadcasts, an observation corroborated by a Pew Research Center survey in

which 59 percent of respondents reported “regularly” watching and an additional

23 percent reported “sometimes” watching local news (2004).3

2The 25 markets in the data, in order of size, are New York, Los Angeles, Chicago,Philadelphia, San Francisco-Oakland-San Jose, Boston, Dallas, Washington, D.C., Atlanta,Detroit, Houston, Seattle-Tacoma, Tampa-St. Petersburg, Minneapolis-St. Paul, Phoenix,Cleveland-Akron, Miami-Ft. Lauderdale, Denver, Sacramento-Stockton-Modesto, Orlando-Daytona Beach-Melbourn, St. Louis, Pittsburgh, Baltimore, Portland, OR, and Indianapolis.

3The study showed no large difference in local news viewing habits by income, educationlevels, or sex. However, older respondents did report watching the news more regularly thanyounger respondents and blacks reported watching more regularly than other racial groups.

5

Table 1: Average Ratings and SharesWeekdays Weekends

Local Start Time Rating Share Rating Share

5:00 a.m. –7:59 a.m. 2.9 13 3.5 138:00 a.m. –10:59 a.m. 2.5 9 4.3 1111:00 a.m.–4:59 p.m. 4.2 13 3.9 105:00 p.m. –8:59 p.m. 6.4 12 5.5 109:00 p.m. –11:59 p.m. 7.9 14 6.7 12

*Shares are reported as whole numbers in accordance with

Nielsen Media Research policy.

Biographical Data

To accompany the ratings data, I collected information from the biographies and

pictures of news teams that nearly every station makes available online and aug-

mented this with information from newspaper reports, news broadcasts, press

releases, and other sources of data on news staff. Using these sources, I made

detailed notes on the characteristics of each of the thousands of on-air employ-

ees in the sample of stations. On-air jobs were divided into 4 categories: news

anchor, sports anchor, weather anchor, and reporter. Each person’s occupation

was noted and the days and times at which the anchors regularly appear were

also collected. Station employees sometimes have more than one of these job

titles. Most commonly, a news anchor for one newscast may also be a general

assignment reporter for others or a sports anchor might also be a sports re-

porter. In these cases, both titles were noted although the more senior position

alone is coded for the purposes of summary statistics.

In addition to position, data were also collected on other characteristics of

the on-air staff. If available, information was collected on origin (state native or

not), first year at the station, first year of paid employment in television news,

total number of stations an employee has worked for, approximate age range

(20–29, 30–39, etc.), and highest degree obtained. Data were also collected on

6

the sex, race, hair color,4 and ethnicity of each employee. Measures of race and

ethnicity are obviously subjective. In general, an employee was considered to

be white and non-Hispanic unless there was direct evidence to the contrary. In

the majority of cases it appears that minorities are members of the various large

national associations for minority journalists such as the National Association of

Black Journalists. Membership in such an organization, mention of the person’s

race or origin, or clear visual evidence were used to decide if a person is black or

Asian rather than white. “Hispanic” is an ethnic identifier rather than a racial

characteristic but, in the interest of simplicity and because nearly all of the

Hispanic journalists in the sample would have also been categorized as “white,”

Hispanics were entered as a separate racial category. As a result, each employee

was classified as white, black, other, or Hispanic. All other races fall into the

category of “other,” which is referred to as “Asian” because of the 125 employees

in this category, 2 were Native American and the remaining 123 were Asian.

Table 2 reports average characteristics of the on-air staff by race. Inter-

estingly, minorities are more likely than whites to be a news anchor, a highly

visible position. Also, fewer minorities, especially Hispanics and Asians, are

male. Minorities also tend to be younger and more educated than their white

counterparts. A final interesting pattern is that Hispanics and Asians tend to

have less experience and tenure than whites and blacks, but to have worked

at a similar number of stations, suggesting that they may change jobs more

frequently.

Station Characteristics

In addition to providing ratings data, Nielsen Media Research also furnished

data on the number of people and households in each DMA as well as breaking4An indicator of whether an employee was blond was created. This is likely just as subjec-

tive a measure as race, but I included it because I was curious to see if there was any evidenceof preference for blonds.

7

Table 2: Average Characteristics of On-Air StaffWhite Black Hispanic Asian

news anchor 26.0 43.3 34.2 36.8weather anchor 15.4 5.8 3.1 3.2sports anchor 7.2 9.2 3.7 3.2reporter 51.7 41.7 59.0 56.8male 64.2 50.4 35.4 22.4age 42.7 41.7 38.3 35.4state native 25.0 19.7 42.9 21.6post-college degree 10.8 18.4 13.0 16.0years at station 9.2 8.7 6.8 4.9years in local television news 18.9 18.7 15.8 12.7number stations worked at 3.6 3.6 3.8 3.8number of observations 1850 381 161 125

this down by age, sex, and black and Hispanic composition. In order to measure

the Asian composition of the markets, data from the 2000 U.S. Census for each

corresponding metropolitan area were used. Therefore, the average characteris-

tics of each market are known, but ratings and shares were only made available

as aggregate numbers and not broken down by these demographic groups. In

addition, local television listings were used to add information on the number

of stations in each DMA that broadcast local news in English, the number of

stations that broadcast local news in Spanish, and a dummy variable indicating

that the DMA has a 24-hour local news station.

Furthermore, the biographical information for each station was aggregated to

calculate average characteristics of each station’s on-air staff such as the percent

who are black and the average tenure at the station. After matching anchors to

time-slots, indicators of the characteristics of the anchor team that is specific

to each broadcast were also included. However, because reporters do not tend

to appear at regular times, they were not matched to specific broadcasts.

Information was also collected from each station’s regional chapter of the

Academy of Television, Arts, and Sciences, which have annual “Emmy Awards”

for their respective regional stations. The awards process requires stations to

8

submit news tapes for consideration for each category in which an award is

granted and the station would like to be considered. The tapes are sent to

another regional chapter (there are twenty in all) for review and winners are

selected. There can be one, more than one, or no winner in a particular category.

Emmy awards are used as a proxy for station quality. While the number of

categories and the level of competition differs by region, fixed effects estimators

which compare stations in the same market are used in later analyses, so this

issue does not present a problem. For 23 of the 25 DMAs, data from the 2003

local Emmy awards were used. For New York City and Detroit, which did not

have 2003 Emmy information available, data from the 2004 Emmy awards were

used.

Of the 100 possible stations (4 per each of the 25 DMAs), 12 were dropped

because they did not have local news programming or little or no biographical

information was available, leaving 2, 569 biographical observations for 88 sta-

tions across 25 markets. These data were combined with the Nielsen ratings

to leave 762 newscast observations. One observation might be the weekday 6

p.m. news on the NBC affiliate in Atlanta, Georgia, and another might be the

weekend 11 p.m. news on another affiliate in the Seattle-Tacoma area. Each of

these observations includes the show’s average rating and share for November

2003 as well as characteristics of the market, demographic characteristics of the

station’s on-air employees and the anchors on that particular show, indicators of

the amount of competitive programming at that time, controls for experience,

and indicators of the quality of the station’s local news programming. Table A.1

in the appendix presents a summary of all the basic variables in the data set.

The average station has about 29 on-air employees, with an average of 9 news

anchors, 3 weather anchors, 2 sports anchors, and 15 reporters. Twenty-six per-

cent of the on-air employees in this sample are minorities. This is somewhat

larger than the 21 percent reported by Papper (2003) in his summary of the

9

Table 3: Proportion of Shows with a Minority AnchorMinority Minority Minority

Time/Week Part News Anchor Weather Anchor Sports Anchor

morning/mid-day 0.61 0.06 0.96evening 0.51 0.13 0.40difference* 0 .10 -0 .06 0 .56weekend 0.47 0.16 0.57weekday 0.59 0.08 0.71difference* -0 .12 0 .08 -0 .14

*All sample proportion differences are significant at the 10% level.

RTDNA/Ball State University survey of news directors, but that could be ex-

plained by differences in samples. Specifically, the sample used here is only the

25 largest markets in the United States while Papper’s results are based on more

markets and different methodology.5 It may simply be the case that stations in

larger markets are more diverse than stations in smaller markets. While there is

a large number of minorities appearing on-air at local news stations, the match

between station composition and market demographics is closer for blacks and

Asians than for Hispanics. Local stations have an average of 15 percent black

employees while the average market is 12 percent black and an average of 5

percent Asian employees while the average market is 5 percent Asian. However,

stations have an average of 6 percent Hispanic employees while the average

market is 14 percent Hispanic. The difference could be an artifact of methodol-

ogy: Hispanics may have been miscategorized more often than the other racial

groups.6

While minority representation is fairly high, it is frequently suggested that

minority journalists are relegated to the lower-rated time slots.7 For each show,5The RTNDA survey conducted by Papper asked station managers what percent of their

staff are minority.6The basic assumption when identifying either group was that a person was white unless

“proven” otherwise. Under this rule, to be classified as Hispanic, a person had to havemore than a Spanish surname; he or she had to be a member of an organization of Hispanicjournalists, mention heritage in their biography, or have some other piece of clear evidence.In the case of blacks and Asians, visual evidence could frequently also be used, making themless likely to be incorrectly identified as white.

7For example, an article in the St. Petersburg Times reports that some minority journalists

10

indicators were constructed for the presence of at least one minority (black,

Hispanic, or Asian) news anchor, a minority weather anchor, and a minority

sports anchor. Table 3 summarizes the proportion of shows that have a mi-

nority anchor by time (morning/mid-day or evening) and part of week. The

sample proportions suggest that morning news shows are 10 percent more likely

to have at least one minority anchor than are evening newscasts, but that week-

end newscasts are 12 percent less likely than weekday casts to have a minority

anchor. This trend is reversed for minority weather anchors; weekend shows are

more likely to have minority weather anchors, but so are evening shows. The

results for news and weather anchors, therefore, do not offer consistent evidence

that minorities are segregated to lower-rated slots. However morning and mid-

day shows are 56 percent more likely and weekend shows are 14 percent less

likely to have a minority sports anchor. This trend appears similar to that for

news anchors, but, because many local sporting events take place on weekends,

it is possible that the weekend sportscasts are actually more popular than week-

day sportscasts.8 If this is the case, then it would appear that minority sports

anchors tend to be segregated to lower-rated slots.

While there is no conclusive evidence that minority anchors are segregated

to lower-rated time slots, the possibility of such endogeneity in the relationship

between race and ratings must be addressed. In order to measure the presence

and magnitude of customer racial preferences, it is necessary to see the impact

on ratings of an exogenous change in race. If the race of anchors is also af-

fected by the rating a time slot can generally be expected to get, then estimates

of customer discrimination against minorities could be biased upward. This

possibility will be addressed in the econometric model.

refer to weekends as the “weekend ghetto” (Deggans, 2003).8The assertion that weekend sportscasts may be more popular than weekday sportscasts

is also supported by station scheduling of sports news and commentary. Local stations thathave special news sections or news shows devoted to sports almost universally broadcast theadditional coverage on weekends rather than weekdays.

11

Turning to the market as a whole, the data indicate that there is quite a

bit of competition in the local news market relative, at least, to sports markets

where there is frequently only one home team or only one game being televised

on broadcast television. The average market has 6 stations that broadcast local

news in English and 1 station broadcasting local news in Spanish, and about a

third of the cities have a 24-hour local news station. Additionally, at the time

of any given news broadcast there are an average of 3 broadcasts in English on

the air (not including any broadcasts on 24-hour channels) and 0.3 broadcasts

in Spanish. The presence of competition presents an opportunity to examine

the strategic response of firms to the actions of their rivals.

So far I have shown that stations tend to have a quite a few minorities on-air

and that there is not strong evidence that these minorities are relegated to less

visible positions or time-slots. However, these aggregate statistics mask impor-

tant intra-market differences. Table 4 reports the average differences between

the largest and smallest station composition variables in a DMA and tests the

hypothesis that this difference is equal to 0. For each of the four measures, the

null can strongly be rejected.9 On average, in each DMA the station with the

highest number of minorities has 4 more minorities on staff than the station

with the lowest. Similarly, the station with the highest minority composition

(pctminority) has 12 percentage points more minorities than the station with

the lowest minority composition. This indicates that the stations within a mar-

ket area do not look alike: one station has significantly more minorities than

another. Stations may be differentiating along racial lines, but there is nothing9This test of the difference in sample proportions relies on a normal approximation of

the binomial distribution. While it presents a simple way to illustrate the variation withinmarkets, the actual sample sizes here are small enough that the law of large numbers on whichthe test relies does not hold. As an alternative, I assumed that the probability of hiring ablack (Hispanic) at each station was equal to the percent of blacks (Hispanics) in the market.For each market, I then calculated the probability of seeing both a station with the minimumobserved composition and a station with the maximum observed composition. Averagingthe results across markets, this probability was 5.5 percent for blacks and 6.7 percent forHispanics. So the results still indicate a significant difference between station composition.

12

Table 4: Tests of Within-DMA Station DifferencesH0 : D sd t-stat P > t

max(no minorities)−min(no minorities) = 0 3.80 2.48 7.65 < 0.001max(no black)−min(no black) = 0 2.32 1.41 8.25 < 0.001max(no hisp)−min(no hisp) = 0 1.64 1.66 4.95 < 0.001max(pctminority)−min(pctminority) = 0 11.98 7.46 8.03 < 0.001

in the existing customer discrimination theory that predicts or explains why

this might happen.

3 Theoretical model of racial differentiation inresponse to customer discrimination

Basic model

Becker’s (1957) well-known model of customer discrimination assumes that cus-

tomers will interact with one employee when they purchase the output of a firm

and that they act as though the price is marked up if that employee is black. He

shows that, in equilibrium, firms will pay a lower wage to black employees and

charge a lower nominal price for their output than for that of white employees.

There is no implication for overall firm composition. In equilibrium, employers

are indifferent between black and white employees; black employees have a lower

marginal revenue product, but also have a lower marginal cost.

However, the assumptions that support this prediction are not always realis-

tic. In many markets, local news being one, customers cannot interact with only

a black or only a white employee. Rather, a customer will come into contact

with many of a firm’s employees. Incorporating this possibility into Becker’s

theoretical model, and using some of the tools of models of product differen-

tiation, we will see that firms may, in fact, have a preference for the racial

composition of their workforces.

Suppose that there are two possible groups from which a firm can hire em-

13

ployees: minorities and non-minorities. Let m ∈ [0, 1] be the percentage of a

firm’s workforce that is composed of minorities. There are N consumers who

each purchase one unit of output and have preferences over the racial composi-

tion of a firm so that consumer i behaves as if he paying P + dim for that unit

when he buys it for price P from a firm with racial composition m. The discrim-

ination coefficient, di, is distributed uniformly over [L, H]. Note in particular

that if L < 0 some consumers have a preference for interacting with minorities.

Assume that there are two firms in this market and that each can choose

the racial composition of its workforce and the price that it will charge for

its output. This set-up is quite similar to that of a Hotelling-type model of

product differentiation and will be solved as a two-stage game in which the firms

first choose racial composition simultaneously and then, given the composition

choices, choose prices.10 To solve for the Nash equilibrium, we work in reverse:

we first see what the equilibrium prices are for fixed composition and then solve

for the optimal composition given these prices.

Assume, then, that Firm 1 and Firm 2 have fixed racial compositions m1

and m2, respectively, and, without loss of generality, that m1 < m2. Then

consumer i will buy from Firm 1 if p1 + dim1 < p2 + dim2. Relying on the

uniform distribution of di, we can derive the demand for output from each firm:

D1(p1, p2) =(

N

H − L

)(H − p2 − p1

m1 −m2

)(1)

and10There is, however, a key difference between the model presented here and a standard

model of product differentiation. In the standard set-up, the distribution of customer tasteslies in the same space on which the firms locate. However, rather than assuming that eachconsumer has an ideal employee racial composition and that there is a uniform distribution ofthis ideal, I assume that consumers are either prejudiced (di > 0) or not prejudiced (di < 0)and that it is the intensity of this preference that varies. I find this specification to be moreplausible as well as in keeping with previous work on discrimination. Moreover, this model’sprediction of differentiation holds for a wide variety of functional forms for the consumer’s pricefunction. A similar prediction is found with a standard differentiation model with quadratictransportation costs but, as has been demonstrated, the results are sensitive to the functionalform of costs.

14

D2(p1, p2) =(

N

H − L

)(p2 − p1

m1 −m2− L

). (2)

Each firm faces the same constant cost c per unit of output. Then Firm i

chooses its price to maximize total profits given by

Πi = (pi − c)Di(pi, p−i). (3)

Maximizing this function for each firm and solving for the Nash equilibrium

prices, we obtain

p∗1(m1,m2) = c +13(L− 2H)(m1 −m2) (4)

and

p∗2(m1, m2) = c +13(2L−H)(m1 −m2). (5)

Moving to the choice of composition given the optimal prices, each firm will

choose m to maximize the reduced-form profit function

Πi(m1,m2) = [p∗i (m1,m2)− c]Di[m1,m2, p∗1(m1,m2), p∗2(m1,m2)]. (6)

Noting that in the Stage 1 maximization problem each firm sets Di + (pi −c)∂Di/∂pi = 0 and using the envelope theorem,

dΠ1

dm1= (p∗1 − c)

(∂D1

∂m1+

∂D1

∂p2

∂p2

∂m1

)(7)

and

dΠ2

dm2= (p∗2 − c)

(∂D2

∂m2+

∂D2

∂p1

∂p1

∂m2

). (8)

Solving the two maximization problems,

dΠ1

dm1=

N

9

[(L− 2H)2

L−H

]< 0 (9)

15

anddΠ2

dm2=

N

9

[(2L−H)2

H − L

]> 0. (10)

So, in equilibrium, m∗1 = 0 and m∗

2 = 1; that is, Firm 1 will always choose to

have a workforce that is all non-minority while Firm 2 will be made up solely of

minorities. The equilibrium price differential will depend on the distribution of

tastes. If L > 0, then Firm 2, which is all minority, will charge a lower price than

Firm 1, which is all white. However, if L and H are sufficiently low (H < −L)

so that there is a relatively large number of consumers with preferences for

minorities relative to those who are prejudiced against them, Firm 2 will charge

a higher price than Firm 1.

This model of racial differentiation has implications for what we might see

empirically. Firms will want to differentiate by racial composition in order to

gain control of certain segments of the market. The equilibrium relationship be-

tween racial composition and demand depends on the distribution of consumer

preferences. In the case where H < −L, that is in which there are more con-

sumers who prefer minorities than do not, firms with low minority composition

will charge less and have lower demand than their rivals. But if H > −L, that is

if more consumers are prejudiced against minorities than prefer them, then firms

with low minority composition will cater to the most prejudiced consumers and

will charge more and have greater demand than their competitors. Moreover,

the response of demand to changes in firm composition would be different for

the different firm types. For instance, in the case where all consumers are prej-

udiced, we could see a more sharp decline in demand in response to an increase

in minorities for low-minority firms than for high minority firms which cater to

the less-prejudiced customers.

Incorporating costs of altering racial composition

One potential shortcoming of this theoretical model is that the prediction of

16

maximal product differentiation seems extreme. Given the presence of equal

employment opportunity (EEO) laws and racial differentials in labor supply,

it may be quite expensive for firms to differentiate completely. To incorporate

this possibility, suppose that Firm i incurs an additional cost of altering its

racial composition from arbitrary level R which might be, for instance, the

racial composition of the market as a whole. This cost could be the result of

increasing the risk of lawsuits under EEO laws or of increasing search costs as

a firm attempts to find minorities to fill all of its slots. Given quadratic costs

of adjusting racial composition, Firm i’s profit function is

Πi = (pi − c)Di(pi, p−i)− (mi −R)2. (11)

This cost of altering racial composition does not affect the demand functions

or firm i’s best response price function, but does alter its location decision. Now

the Nash equilibrium compositions are:

m∗1 = max

[0, R− 1

18

(N

H − L

)(2H − L)2

]∈ [0, 1] (12)

and

m∗2 = min

[R +

118

(N

H − L

)(2L−H)2, 1

]∈ [0, 1]. (13)

So, if there is a cost of adjusting composition, the firms will differentiate, but

not necessarily completely. The equilibrium distance between them depends on

the size of the population and the tastes of consumers. Note that

dm1∗dN

= − 118

(2H − L)2

H − L≤ 0 (14)

and

dm2∗dN

=118

(2L−H)2

H − L≥ 0 (15)

17

So, as the population size, N , grows, the degree of separation increases. How-

ever, the comparative statics for consumer preferences are not straightforward.

The effect of changes in L and H on product differentiation depends on the size

and sign of L and H.

4 Analysis of Station Composition and CustomerPreferences

I will first look at firm racial composition to see if the characteristics of competi-

tors are negatively related, and I will then look at the effect of these characteris-

tics on demand to see if it is consistent with the model of product differentiation.

Racial Composition of Employees

The theoretical model presented in the preceding section suggests that a sta-

tion’s racial composition is negatively correlated with the composition of the

other stations in the market. But it can also depend on employer discrimi-

nation and the demographic characteristics of the market. The characteristics

of the market are not so important as a measure of the potential labor pool,

but rather because of how they interact with EEO laws.11 While the rules and

enforcement of EEO laws in the local television industry have changed over

previous decades, since their enactment in 1967, station racial composition has

generally been compared to the composition of the surrounding market.

In order to analyze the determinants of employee composition, I use the

station-level data to estimate random effects regressions of racial and gender

composition (pctblack, pcthispanic, pctasian, pctfemale) and age composition

(age, the average age of on-air employees at a station) that allow for possible

correlation of the error terms within markets. Explanatory variables include11Journalists appear to move easily and frequently between markets. In this sample, the

average journalist has worked at 4 stations and at the average station only a quarter ofemployees are natives of that state.

18

station and market characteristics.12

A potential concern is that the right-hand-side variable measuring the av-

erage composition of competing stations may be correlated with unobserved

market characteristics. Consider, for instance, the regression for station black

composition:

pctblack = α1 + dmapctblackα2 + (other stat pctblack)α3 + Xα4 + ε (16)

where X contains additional station and market characteristics listed in Ta-

ble 5. The theoretical model predicts that α3, the coefficient on the average

composition of competing stations, should be negative. However, any unob-

served market characteristics that influence the composition of other stations

should also influence the composition of the observed station. For instance, sup-

pose that stations in more educated markets are more likely to hire minorities.

Because education is not observed, α3 will be biased upward. In fact, any un-

observed market characteristic that influences station composition should have

a similar effect for all stations. Therefore, to the extent that negative correla-

tion is observed in the presence of unobserved market characteristics, this only

bolsters the results.

However, an endogeneity problem remains due to the fact that racial compo-

sition regressions represent a system for each market in which the composition

of station i is determined by the average composition of the remaining stations

in the market. The appendix presents a brief derivation of the statistical prop-

erties of the estimator of α3 in such a system. In the case of positive correlation12A negative binomial count model for the actual number of employees from any one group

was also estimated as was a truncated dependent variable model. Both yielded similar results.Another possibility is that there is substitution between worker types. That would suggest asystem of equations with endogenous dependent variables, but no instrument suggests itselffor identifying these variables. However, in that case, omitting measures of the station com-position from the right hand side as done here could lead to bias. If, for instance, black andHispanic workers are substitutes, then anything that tends to increase demand for one groupwill decrease the demand for the other group. This possibility should be kept in mind wheninterpreting the results.

19

between the error terms (as seems likely), the probability of rejection for a neg-

ative coefficient estimate is not overstated. However, it is possible that if the

relationship between station compositions is positive and very large, the ex-

pected value of the estimate could be negative. For this reason, I proceed with

the racial composition regressions but note that, while a significant negative

estimate is likely to indicate a true negative correlation between composition,

the magnitude of the effect will be biased.

Table 5 reports the results of the five regressions. There is not much evidence

of discrimination by employers. An indicator that a station has a minority

station manager and/or news director, minmanager, was included in order to

see if the race of managers has any relationship with the race of employees. For

instance, white managers may discriminate against black employees while black

managers may not.13 However, the coefficient is positive and significant only in

the sex regression, indicating that minority managers are positively correlated

with female employees. An indicator that a station has a female manager was

also included, but the coefficient is not significant in any of the regressions.14

It is possible that station managers and news directors do not bear all of the

burden of hiring decisions. But in this case, one would expect that there would

be a positive correlation between the characteristics of managers and on-air staff

because both are employees subject to the same employer discrimination. That

there is not strong evidence of this only further supports the conclusion that

employer discrimination does not play a role

13Antonovics and Knight (2004), for example, use evidence from traffic stops in Boston toshow that police officers are more likely to conduct a search if the driver is of another race,suggesting that racial preferences vary by race.

14Because it was especially difficult to collect information on the characteristics of stationmanagers, the race of the station manager or news director is only known for 49 of the88 stations. Hence, the coefficient on minmanager is actually comparing the impact of aminority manager relative to a white manager or to a manager of unknown race. This couldbe the source of the lack of significance in the remaining 3 regressions. However, the resultsare similar if a dummy indicating that manager race was not observed is included. Becausemanager names were more readily available than biographies and pictures, the sex of managerswas identified at 79 of the 88 stations.

20

Tab

le5:

Ran

dom

Effec

tsR

acia

lC

ompos

itio

nR

egre

ssio

ns

dep

.va

riabl

es:

station

perc

entbl

ack

,H

ispa

nic

,A

sian

and

fem

ale

and

mea

nage

pctb

lack

pcth

ispanic

pcta

sian

pctf

em

ale

age

coef

seco

efse

coef

seco

efse

coef

se

const

ant

11.3

0118.0

7-1

08.3

272.2

647.7

983.2

1-4

12.6

9144.9

946.4

167.6

5M

ark

etC

hars

.dm

apct

bla

ck1.0

70.2

5<

0.0

10.1

50.0

90.1

8-0

.54

0.3

10.0

10.1

4dm

apct

his

panic

-0.1

80.2

30. 6

50.1

4-0

.01

0.1

61. 0

10.2

8-0

. 36

0.1

3dm

apct

asi

an

0.1

90.3

0-0

.28

0.1

80.6

80.2

10.4

30.3

7-0

.22

0.1

7dm

apct

fem

ale

-1.0

31.9

01.8

01.1

6-1

.60

1.3

38.0

52.3

3-0

.68

1.0

9dm

apct

young

0.4

10.5

6-0

.25

0.3

4-0

.25

0.3

9-1

.79

0.6

80.4

50.3

2dm

apct

pri

me

0.0

80.5

70.9

30.3

5-0

.22

0.4

01.0

10.7

00.1

30.3

3lo

g(p

erso

ns)

-0.3

52.3

4-2

.84

1.4

33.1

61.6

5-0

.32

2.8

80.5

91.3

4so

uth

0.8

82.7

71.0

81.7

01.4

61.9

56.0

53.4

0-0

.32

1.5

9w

est

-4.5

03.7

3-3

.63

2.2

83.0

32.6

33.0

34.5

83.3

22.1

4m

idw

est

4.7

92.2

92.7

01.4

01.4

31.6

14.8

22.8

1-1

.39

1.3

1en

glish

stati

ons

0.3

41.1

52.2

10.7

00.1

50.8

1-1

.67

1.4

10.7

40.6

6sp

anis

hst

ati

ons

0.8

92.0

03.5

01.2

2-1

.50

1.4

1-1

0.5

12.4

63.0

31.1

524hr

stati

ons

-1.4

41.7

5-1

.64

1.0

70.0

21.2

31.4

82.1

51.3

11.0

0N

BC

1.8

31.5

10.3

50.9

31.1

71.0

72.0

81.8

51.3

90.8

7C

BS

2.1

41.5

50.6

00.9

50.6

61.0

93.1

61.9

01.8

50.8

9A

BC

2.4

11.5

80.3

80.9

7-0

.06

1.1

12.1

01.9

41.8

00.9

0Sta

tion

Chars

.m

inori

tym

anager

2.6

51.7

80.1

31.0

9-0

.94

1.2

56.6

42.1

8-0

.68

1.0

2fe

male

manager

1.4

81.2

60.4

10.7

7-0

.82

0.8

92.1

41.5

4-0

.16

0.7

2oth

erst

at

pct

bla

ck-0

.52

0.2

3-0

.25

0.1

4-0

.16

0.1

60.2

60.2

80.2

40.1

3oth

erst

at

pct

his

panic

-0.3

80.3

9-1

.14

0.2

40.0

30.2

80.3

40.4

80.0

80.2

2oth

erst

at

pct

asi

an

-0.0

90.2

90.1

10.1

8-0

.14

0.2

00.6

40.3

5-0

.01

0.1

6oth

erst

at

pct

fem

ale

0.1

20.1

60.0

70.1

00.0

80.1

2-0

.89

0.2

00.2

70.0

9oth

erst

at

age

0.9

20.3

50.2

10.2

10.0

40.2

51.6

20.4

3-0

.33

0.2

0

no.

obse

rvati

ons

88

88

88

88

88

*Bold

coeffi

cien

tsare

sign

ifica

ntat10%

leve

l.

21

in this market.15

As suggested by the theoretical model, there is a significant negative re-

lationship between the racial compositions of stations within a market. A 1

percentage point increase in the average percent black at other stations in a

market is correlated with a 0.5 percentage point decrease in a station’s own per-

cent black. And a 1 percentage point increase in the average percent Hispanic

at other stations is correlated with a decrease of 1.1 percentage points in own

percent Hispanic. Similar trends hold for sex and age. A 1 percentage point

increase in the average percent female at competing stations is correlated with

a 0.9 percentage point decrease in percent female. And a one year increase in

the average age of employees at competing stations is correlated with an own

average age decrease of four months. While there is no evidence of this sort-

ing for Asian employees, this is also the only one of the categories in which

the market composition is proxied with Census data on the metropolitan area

because Nielsen Media Research did not provide information on Asian market

composition.

The negative correlation between the representation of a group at a station

and at its competitors suggests that there is some intra-market sorting going on

among competing stations. In some markets, this might be the result of a small

pool of potential employees, so that if a firm hires a black employee, the ability of

its competitors to hire one is reduced. However, as mentioned before, television

journalists move frequently between stations and regions, suggesting that even

if a particular market does not have many minorities, that does not mean that

there are no minorities available to hire. Furthermore, with its reputation as a

tough field to crack, there appears to be no shortage of potential journalists of15Another way to look for employer discrimination is to control for the corporate owner

of a station. There are 23 corporate owners (such as News Corp., Gannett, Disney, etc.) inthis sample. However, coefficients on owner indicators were neither individually nor jointlysignificant.

22

any race. Hence, it seems likely that results are indicative of racial differentiation

by firms.

While the evidence supports sorting among stations, there is also a signif-

icant positive relationship between the racial composition of a market and the

composition of all stations within it. A 1 percentage point increase in the per-

cent of a market’s population that is black is correlated with a 1.1 percentage

point increase in the percent black at a station. Similarly, a 1 percentage point

increase in the percent of a market’s population that is Hispanic is correlated

with a 0.7 percentage point increase in the percent Hispanic at a station. There

do not, however, appear to be cross-racial effects for markets and stations; the

percent black in a market is not significantly related to the percent Hispanic

at a station and vice versa. Looking at females, there is a large effect of the

gender composition of a city on the composition of a station, an especially sur-

prising result given the low variance in dmapctfemale, which only ranges from

50 to 53 percent in this sample. With the exception of age, therefore, it seems

that the characteristics of television journalists are related to the corresponding

characteristics of the market.

As previously mentioned, it is not clear to what extent this finding might

represent the effects of labor supply, station response to audience demand, and

the impact of EEO laws. In particular, because of their enforcement by the

Federal Communications Commission (FCC), EEO laws may play an especially

large role in the hiring decisions of local stations. In order to examine this

possibility, I take advantage of several shifts in the provisions and enforcement

of these laws by the FCC.

EEO laws actually encompass a range of congressional acts pertaining to job

discrimination. The first and most well known is Title VII of the Civil Rights

Act of 1964, which forbids employers from discriminating based on a variety

of characteristics. In response to this act, the FCC, which oversees the federal

23

Table 6: EEO Enforcement RegimesRegime Time Period Enforcement

Regime 1 1970-1986 Beginning of FCC regulation: Stations submit writtenEEO reports to FCC. Stations that did not employ suf-ficient numbers of minorities relative to their marketwere examined closely at the time of license renewal.

Regime 2 1987-1997 Stricter regulations: FCC shifts emphasis to minorityoutreach and recruitment. Stations must submit addi-tional forms detailing these efforts and results continueto be tied to license renewal

Regime 3 1998-1999 No regulations: FCC practices are ruled unconstitu-tional in 1998 and are suspended.

Regime 4 2000-2003 Fluctuating regulations: FCC tries two sets of rulesthat tie EEO compliance to licensing.

licensing of local stations, began requiring stations to submit written reports

summarizing the composition of their employees in 1970. Stations that had 50

or more employees or that did not have sufficient minorities on staff relative to

the composition of their market were reviewed by the FCC at licensing renewal

time. In 1987 the FCC modified its policy to place greater weight on recruit-

ment efforts and, as a result, the amount of required paperwork and monitoring

increased. However, in 1998 the tying of federal licensing to these requirements

was ruled unconstitutional and the FCC suspended its EEO program, leading to

a two year period in which no FCC enforcement was in place. In 2000 the FCC

implemented a new standard but this was also quickly ruled unconstitutional,

leading to another period during which the FCC did not attempt to mandate

or enforce adherence to EEO guidelines. In 2003 the FCC again promulgated

regulations that required stations to submit information on their minority re-

cruitment efforts and, again, compliance was tied to license renewal. These

changes present several natural “regimes” of EEO laws, summarized in Table 6.

While I do not have data on station racial composition over time, I do

know when most current employees were hired. In order to get an idea of the

importance of FCC enforcement of EEO laws, I estimate a set of regressions of

24

the percent of employees who are black and hired during a particular regime

(pct(black ∗ regimej=i)) on market racial composition. The regression is of the

form

pct(black ∗ regimej=i) = dmapctblackβ1 + pct(black ∗ regimej 6=i)β2 + ε. (17)

β1 measures the relationship between the market composition in 2003 and sta-

tion composition during each regime. β2 is included to attempt to minimize

the impact of these regressions being based only on employees who were both

hired during a particular regime and are presently at the station. For instance,

the percent of employees who are black and who were hired before 1987 could

be low because of attrition. In that event, there would be more blacks hired

during later regimes to replace those lost previously, which would be picked up

by the second coefficient, preventing the impact of the city racial composition

from being biased downward.

Table 7 reports the estimates for β1 for regressions where the dependent vari-

able is percent black during each of the 4 regimes, as well as separate regressions

for Hispanics and Asians.16 Looking at the results for blacks, the impact of the

racial composition of a market on the hiring practices of a station increases

between regimes 1 and 2 as FCC enforcement becomes stricter, decreases with

Regime 3 and the period of lax enforcement, and increases again with Regime

4 and the implementation of new rules. The R2 for each of the regressions

for blacks also increases with stricter regimes. A similar trend is evident for

Hispanics and Asians, suggesting that market composition has more power in

explaining station composition during periods of stricter EEO regulations and

enforcement. This evidence, therefore, does indicate that FCC enforcement of

EEO regulations in the local television market has had an impact on station

hiring practices. However, even during recent periods of confusing or absent en-16Only 2 blacks and no Hispanics or Asians were hired by their station before 1970, so the

analysis does not include the period before 1970

25

Table 7: Impact of DMA Composition on Minority Hiring over TimeBlacks Hispanics Asians

dma dma dmadependent variable: pctblack R2 pcthispanic R2 pctasian R2

pct(regime1 ∗ race) 0.26 0.14 0.03 0.09 0.05 0.06pct(regime2 ∗ race) 0.43 0.32 0.20 0.45 0.31 0.24pct(regime3 ∗ race) 0.21 0.20 0.12 0.31 0.20 0.15pct(regime4 ∗ race) 0.68 0.40 0.25 0.50 0.43 0.23

*bold coefficients are significant at 10% level

forcement, there is still a significant and positive relationship between the racial

composition of a city and minority hiring at stations, indicating that while it

does place some constraint on stations, EEO enforcement is not the only driving

force behind minority hiring.

Customer Preferences

Ratings regressions measure the effect of employee characteristics on customer

demand.17 Regressions are of the form

ratingijt = α + Sijβ1 + N ijtβ2 + υijt, (18)

where ratingijt is the rating of show i in market j at time-slot t, Sij is a vector of

containing the characteristics of the station, and N ijt is a vector containing the

characteristics of that particular newscast. It is plausible and indeed, Hausman

specification tests indicate that there are certain unobserved effects that are

particular to a time slot in a given market and are correlated with the regres-

sors. For instance, if minorities are indeed assigned to slots that are expected

to receive low ratings, then the estimated effect of racial composition may be

downward biased. In an attempt to account for this possibility, the specification

incorporates market and time fixed effects. In other words, the estimates are17Another option would be to use share as the dependent variable rather than rating.

Because share measures viewership only as a fraction of those watching TV, this would es-sentially measure preferences conditional on turning the television on at all, while rating isunconditional on this. If, for instance, customers are prejudiced against blacks and tend notto watch television when blacks are on, the effect of race may be underestimated. In thisinstance, the results of regressions using share are only a little smaller than the results fromthe ratings regressions.

26

based on differences in outcomes for newscasts in the same market at the same

time of day (e.g., early morning weekdays, etc.).

Table 8 reports the results for the above model using a matching speci-

fication to measure station composition. The racial and gender composition

of a station is measured relative to the same characteristics of the market.18

These weighted variables can be thought of as matching measures; the variables

blackmatch, hispanicmatch, asianmatch and femalematch are the ratios of

station composition to market composition multiplied by 100. So, if blackmatch

is equal to 100, the station exactly matches the black composition of the city

while if blackmatch is less than 100 the station under-represents blacks and if

blackmatch is greater than 100 the station over-represents blacks. Using the

matching specification allows for the possibility that the effect of adding blacks

is different in markets with different racial compositions. In addition to these

variables measuring the overall composition of journalists at a station, there are

also indicators for the sex and minority status of the news, weather, and sports

anchors on the particular newscast being observed.19

Turning to the results, only 2 of the 6 anchor variables, which are matched to

each newscast, are significant. The significant estimates suggest that if a male

news anchor is replaced with a female one, ratings decline by 0.61 percentage

points and if a white weather anchor is replaced with a minority one, ratings

decline by 0.51 percentage points. The insignificance of the remaining anchor

variables could indicate that the effect of the specific anchors on any one show

is not large; or, it could be a product of the small sample size.

The variables measuring overall station composition, however, do suggest

that customers have preferences for employee characteristics over all of the di-18Age is not weighted because Nielsen Media Research did not provide detailed age charac-

teristics of markets19More detailed indicators for anchor team composition were also considered so that,for

instance, a solo white male anchor could be compared to a black male/white female duo andso on. However, these more detailed indicators were jointly insignificant.

27

Table 8: Fixed Effects Ratings Regressionsdep. variable: average Nielsen rating for November 2003

coef se coef se

Employee Characteristics Anchor Characteristicspctnative -0.021 0.009 sport anch 1.164 0.287pcthighed -0.019 0.013 duo -0.203 0.237tenure 0.369 0.062 trio 1.1x10−4 0.504experience -0.015 0.029 min. anchor 0.178 0.156no. stations 0.633 0.301 female anchor -0.610 0.356pctblond 0.122 0.019 min. weather -0.507 0.247age -18.380 9.245 female weather -0.236 0.173age2 0.434 0.224 min. sports 0.229 0.236age3 -0.003 0.002 female sports -0.111 0.527female match 1.149 0.295 Station Characteristicsfemale match2 -0.015 0.004 NBC 2.305 0.256female match3 6.4x10−5 1.8X10−5 CBS 1.468 0.287black match -0.118 0.018 ABC 2.279 0.280black match2 0.001 1.2X10−4 length 0.254 0.054black match3 -1.74x10−6 2.48X10−7 pct emmys 0.024 0.004hispanic match 0.052 0.015 post network -0.823 0.211hispanic match2 -3.9x10−4 2.2x10−4 pre network 0.305 0.188hispanic match3 -2.3x10−8 8.0x10−7 no. persons 0.011 0.021asian match 0.018 0.005 min. manager -0.544 0.317asian match2 7.1x10−5 2.6x10−5 female manager 0.469 0.207asian match3 8.5x10−8 3.3x10−8 constant 228.363 125.662

*bold coefficients are significant at 10% level

28

mensions examined. The coefficient on pctblond suggests that a 1 percentage

point increase in the percent of employees who are blond yields ratings that are

0.1 percentage points higher.20 Because specification tests indicate that cubic

functions are appropriate for the remaining composition variables, the match-

ing variables for average station characteristics were entered along with their

squares and cubes. Figure 1 shows the estimated mean of a show’s rating con-

ditional on the five composition variables.21 Looking at the graph for the effect

of the match between station percent black and market percent black, ratings

initially decline as black representation at a station increases but, as station

composition begins to exceed market composition, ratings develop a positive

relationship with black composition. A similar result holds for age, although

it appears that the negative effect of age diminishes as average age increases

but does not exhibit a strong positive trend. For gender composition, there is a

negative but decreasing effect of adding women. The shapes of these functions

are not consistent with Becker’s model of customer discrimination in which di-

minishing marginal utility would cause the negative impact of race to increase

with minority representation. They are also not explained by a taste for diver-

sity, since this would suggest that the functions would be concave. However,

the decline in the negative response as composition increases is consistent with

the theory of station differentiation in which stations sort in response to cus-

tomer prejudices against blacks, women, and older workers, yielding p1 < p2.

The results for blacks, for example, suggest that those stations with a relatively

low composition of blacks are catering to consumers with a high discrimination

coefficient, while those stations with more blacks are catering to consumers who

are without such strong prejudice against blacks or even have preferences for20Interacting hair color with sex did not indicate that there was a significant difference in

how customers respond to blond women relative to blond men. However, the sample of blondmen was much smaller.

21The functions are evaluated over the 10th to 90th percentiles of the independent variablesthat are observed in the data.

29

them. However, if the black composition variable becomes high enough, ratings

will begin to decline again, suggesting that at very high levels of representation

(when there are more than 2 and a half times as many blacks on the station than

in the market, which is only observed for a few outlying stations and hence not

shown on the domain of the graph), even the less prejudiced consumers begin

to exhibit a strong response.

Turning to the remaining two graphs in Figure 1, the results indicate that

at least some consumers have a preference for Hispanics and Asians. Looking at

Hispanics, ratings increase over low levels of composition but decline with the

addition of more Hispanic employees when hispanicmatch exceeds 60. Similarly

for Asians, ratings increase with Asian representation among employees until

that representation exceeds Asian composition in the market. The concavity of

these functions is not consistent with the model of differentiation, but instead

could suggest a taste for diversity in which consumers prefer to see about half as

many Hispanics and about the same number of Asians on television as are in the

market. However, as mentioned earlier, Hispanics may have been categorized

incorrectly more often than other groups. If, for instance, Hispanics employees

in markets with fewer Hispanics were less likely to acknowledge or point out

their ethnicity than were employees in more Hispanic markets, this could bias

the estimates. Finally, stations that broadcast in Spanish or other languages

are not included in this sample. It could simply be the case that these stations

cater to customers with the greatest taste for seeing Hispanic or Asian minority

groups.

In addition to measures of the characteristics of the on-air staff, indicators

for a female and minority manager were also included in the ratings regressions

as a sort of check on the results. Assuming that managers are much less visi-

ble to customers than the on-air journalists, their race or sex should not have

an effect on ratings. However, the estimates suggest a positive and significant

30

Figure 1: Effects of Employee Compositions on Ratings

31

effect for female managers and a negative and significant effect for minority

managers. If these managers were visible, the results could indicate some pref-

erence of customers. An alternative explanation might be that there is some

fundamental difference in the productivity of the two groups in these positions.

Or perhaps these two groups are assigned to different types of stations, so there

is endogeneity here. In either event, it is possible that the results for the on-

air cast members could also be picking up some effect other than strict racial

preferences.

5 Conclusion

Understanding the role and nature of customer discrimination is an important

component of understanding differential outcomes in labor markets because,

to the extent that customer preferences drive outcomes, competition will not

necessarily eliminate discrimination. However, empirical studies have tended

to focus on a single market–that for sports– and, even within this somewhat

limited arena, have provided a variety of findings.

This chapter presents a look at another labor market and a different group

of employees in order to infer the preferences of customers there. While the

initial motivation was driven by the observation of a high number of minorities

on local television news and by an attempt to see if customers might have pref-

erences for some level of diversity, a closer examination of the local news market

provided evidence of a previously unconsidered phenomenon on the employer

side: local stations appear to be sorting along racial lines. While very few tele-

vision stations are all white (most likely as a result of FCC enforcement of EEO

laws), there is significant variation in racial composition within a market that is

masked by averages. The theoretical model presented in Section 3 demonstrates

that if customers have a distribution of preferences for the composition of firms,

32

firms will compete through racial differentiation.

The empirical findings support this model and suggest another layer of com-

plexity to consider with the basic Becker model of customer discrimination.

Local news stations appear to respond to the racial composition of their com-

petitors and to try to differentiate themselves by race, age, and sex of their

on-air employees. For three of the five groups examined, the ratings regressions

indicate that the response of consumers varies with the racial composition of the

firms in a manner consistent with the predictions of this model of racial differen-

tiation. Viewers of the more “white” stations have a stronger negative reaction

to an increase in blacks than do viewers of the “black” stations, suggesting that

stations with few blacks cater to consumers with a high discrimination coeffi-

cient against blacks, while stations with more blacks cater to customers who are

less prejudiced or who prefer blacks. Similarly, viewers of stations with more fe-

males and older employees have a smaller negative response to these groups than

viewers of stations with lower concentrations. While the results for Hispanics

and Asians suggest a customer preference for diversity rather than differentia-

tion or strict racial preferences as originally modeled in the literature, it seems

likely that the exclusion of foreign language local news has biased the estimates.

Taken as a whole, the results here suggest that customer discrimination may be

a more complex phenomenon than we have previously considered.

33

A Variables

Table A.1: List of Variables

Name Description min max mean

Show Characteristicsrating Nielsen rating for November 2003 0.5 17.4 5.1share Nielsen share for November 2003 1 33 12length number of quarter hours that broad-

cast lasts1 12 2.9

post network indicator that broadcast immedi-ately follows national network news

0 1 0.19

pre network indicator that broadcast immedi-ately precedes national networknews

0 1 0.25

weekday indicator that broadcast is on aweekday

0 1 0.76

weekend indicator that broadcast is on aweekend

0 1 0.24

early morn indicator that broadcast begins be-tween 5 a.m. and 7:59 a.m.

0 1 0.33

mid morn indicator that broadcast begins be-tween 8 a.m. and 10:59 a.m.

0 1 0.03

midday indicator that broadcast begins be-tween 11 a.m. and 4:59 p.m.

0 1 0.12

early eve indicator that broadcast begins be-tween 5 p.m. and 5:59 p.m.

0 1 0.29

late eve indicator that broadcast begins be-tween 9 p.m. and 11:59 p.m.

0 1 0.22

no. english casts number of local newscasts in Englishbroadcast during show’s time slot

1 6 3.1

no. spanish casts number of local newscasts in Span-ish broadcast during show’s timeslot

0 3 0.3

sport anch indicator that broadcast has asports anchor

0 1 0.43

minority anchor indicator that broadcast has a mi-nority news anchor

0 1 0.56

female anchor indicator that broadcast has a fe-male news anchor

0 1 0.95

minority weather indicator that broadcast has a mi-nority weather anchor

0 1 0.10

female weather indicator that broadcast has a fe-male weather anchor

0 1 0.24

minority sports indicator that broadcast has a mi-nority sports anchor

0 1 0.68

female sports indicator that broadcast has a fe-male sports anchor

0 1 0.13

Market Characteristics

34

Name Description min max mean

dma rank market rank by size, 1 being largest 1 25 -north indicator that dma is in the north 0 1 0.21south indicator that dma is in the south 0 1 0.35west indicator that dma is in the west 0 1 0.27midwest indicator that dma is in the midwest 0 1 0.17dma pctblack percent of persons in market who

are black2.2 25.7 12.4

dma pcthispanic percent of persons in market whoare hispanic

0.8 42.4 13.8

dma pctasian percent of persons in MSA who areasian

1.1 19.4 4.8

dma pctfemale percent of persons in market whoare female

50.2 53.2 51.9

dma pctyoung percent of persons in market aged2-17

19.4 26.1 23.1

dma pctprime percent of persons in market aged18-49

40.7 50.8 47.1

english stations number of stations that broadcastlocal news in English

4 9 6.2

spanish stations number of stations that broadcastlocal news in Spanish

0 5 1.0

24hr stations indicator that there is a 24-hour lo-cal news channel in market

0 1 0.3

Station CharacteristicsNBC indicator that station is NBC affili-

ate0 1 0.28

ABC indicator that station is ABC affili-ate

0 1 0.24

CBS indicator that station is CBS affili-ate

0 1 0.26

FOX indicator that station is FOX affili-ate

0 1 0.22

no. persons number of on-air employees 11 45 29.2age average age of on-air staff 33.4 48.8 41.5pctnative percent of on-air staff that is area

native0 56.6 25.0

pcthighed percent of on-air staff with educa-tion beyond a bachelor’s degree

0 39.1 11.7

tenure average staff tenure at station, inyears

2 15.4 8.3

experience average staff experience in local tele-vision news, in years

3.0 27.0 17.6

no. stations average number of stations on-airstaff have worked at

2.6 4.7 3.6

pct emmys percent of regional emmys for localnews awarded to station

0 84.6 26.3

pctwhite percent of on-air staff that is white 54.3 100.0 74.2pctblack percent of on-air staff that is black 0.0 38.1 15.0

35

Name Description min max mean

pcthispanic percent of on-air staff that is His-panic

0.0 28.6 5.9

pctasian percent of on-air staff that is asianor other(2 of the 125 observationsare “other”)

0.0 20.0 4.8

pctfemale percent of on-air staff that is female 20.0 57.1 41.8pctblond percent of on-air staff that is blond 0.0 36.8 15.2black match pctblack/dmapctblack 0 322.3 132.0hispanic match pcthispanic/dmapcthispanic 0 193.0 37.3asian match pctasian/dmapctasian 0.0 584.1 113.2female match pctfemale/dmapctfemale 39.3 110.0 80.5min manager indicator that station manager

and/or news director is a minority0.0 1 0.22

female manager indicator that station managerand/or news director is female

0.0 1 0.38

other stat pctblack average pctblack for other stations inDMA

3.1 31.2 15.0

other stat pcthispanic average pcthispanic for other sta-tions in DMA

0.0 24.2 5.9

otherstat pctasian average pctasian for other stationsin DMA

0.0 17.7 4.8

other stat pctfemale average pctfemalefor other station inDMA

30.4 51.5 41.8

other stat age average agefor other station in DMA 34.8 46.5 41.5

*Note that the means for the station characteristics weight each station evenly

and the means for the market characteristics weight each market evenly.

36

B Properties of the Racial Composition Estima-tor

Consider the case in which each market has two competing stations, indexed 1

and 2. The racial composition of each, denoted y, can be modeled as a system

of linear equations:

y1 = y2α + ε1 (B.1)

y2 = y1α + ε2. (B.2)

In this case the reduced form equations are:

y1 =α

(1− α2)ε2 +

1(1− α2)

ε1 (B.3)

y2 =α

(1− α2)ε1 +

1(1− α2)

ε2. (B.4)

In terms of station 1, the estimator α̂ is:

α̂ = α +cov(y2, ε1)

E(y22)

. (B.5)

Assume symmetry so that V ar(ε1) = V ar(ε2) = σ2 and let ρ = cov(ε1,ε2)var(ε1)

. Then

it can be shown that

α̂ = α +(α + ρ)(1− α2)1 + α2 + 2αρ

. (B.6)

If the correlation between the error terms can be eliminated such as in the

random effects model used in the paper, this simplifies to

α̂ =2α

(1 + α2). (B.7)

While the estimator is biased, the sign is expected to be correct. Furthermore,

under the null hypothesis that α = 0, the estimator is consistent. In the event

that correlation remains, ρ is expected to be positive because any unobserved

market factors that influence station composition should influence both stations

37

similarly. Given ρ ≥ 0, then plim(α̂) = ρ under the null and the rejection region

is not overstated.

Extending these results to the case of markets in which there are more firms

present, if there are three firms in a market, then

α̂ = α +(α + 2ρ)(1− α)(2 + α)

2 + α2 + 4αρ + 2ρ. (B.8)

If there are four firms in a market, then

α̂ = α +(α + 3ρ)(1− α)(3 + α)

3 + α2 + 6αρ + 6ρ. (B.9)

Again assuming that the correlation between the error terms is eliminated, the

estimate will now be of the correct sign as long as α < 4 in the case of three firms

and α < 3 in the case of four. In other words, although the estimator is biased,

the estimate is expected to be of the correct sign unless the true relationship is,

in fact, large and positive, which seems unlikely. For both three and four firms,

it is still the case that the estimator is consistent under the null or that, if the

correlation is positive, the rejection region is at least not overstated.

38

References

Kate L. Antonovics and Brian G. Knight. A new look at racial profiling: Evi-dence from the Boston Police Department, 2004. NBER working paper 10634.

Peter Arcidiacono, Jacob Vigdor, and Eric Aldrich. Do people value racialdiversity? Evidence from Nielsen ratings. Working Paper, 2004.

Gary S. Becker. The Economics of Discrimination. University of Chicago Press,1957.

Richard C. Burdekin and Todd L. Idson. Customer preferences, attendence andthe racial structure of professional basketball teams. Applied Economics, 23:179–186, 1991.

Eric Deggans. Minorities fill 2 WFLA weekend anchor jobs. St. PetersburgTimes, May 8, 2003.

James Gwartney and Charles Haworth. Employer costs and discrimination: thecase of baseball. Journal of Political Economy, 82(2):873–881, 1974.

Harry J. Holzer and Keith R. Ihlanfeldt. Customer discrimination and employ-ment outcomes for minority workers. The Quarterly Journal of Economics,113(3):835–867, August 1998.

Lawrence M. Kahn and Peter D. Sherer. Racial differences in professional bas-ketball players’ compensation. Journal of Labor Economics, 6(1), 1988.

Mark T. Kanazawa and Jonas P. Funk. Racial discrimination in professionalbasketball: Evidence from Nielsen ratings. Economic Inquiry, 39(4):599–608,2001.

Clark Nardinelli and Curtis Simon. Customer racial discrimination in the marketfor memorabilia: The case of baseball. The Quarterly Journal of Economics,105(3):575–595, August 1990.

American Society of Newspaper Editors (ASNE). Tables from the 2002 News-room Employment Census, 2003. URL www.asne.org.

Bob Papper. Tables from the RTDNA/Ball State University Survey on Womenand Minorities, 2003. URL www.rtnda.org.

ESPN Research. 2003 audience profile, 2004. URL www.espn.com.

Pew Research Center. 2004 media consumption and reliability survey, 2004.URL www.people-press.org.

Paul M. Sommers and Noel Quinton. Pay and performance in major leaguebaseball: The case of the first family of free agents. Journal of Human Re-sources, 17(3):426–436, 1982.

Eric W. Stone and Ronald S. Warren, Jr. Customer discrimination in profes-sional basketball: evidence from the trading-card market. Applied Economics,31:679–685, 1999.

39